[kvlonge]

Why

When a source fails its freshness check, a scheduled build already reports the blocked models as skipped before execution. The executor still calls the node-start hook for those planner skips, so the worker then reported them as started. The control API rejects a start for a node whose outcome is final ("node change is inconsistent with its current outcome"). The worker stopped, and every node it had not reached ended skipped, including branches independent of the stale source.

Changes

• A worker never reports a start for a node whose outcome it has already reported, such as a planner skip, a skip inherited from a failed sibling, or an unchanged node.
• no_change planner skips need no separate test. The planner never emits a model SKIP with reason no_change: its only model skip is disabled, and blocked models get source_freshness_error or external_upstream_failed. Unchanged models are rebuilt or refreshed. If such an entry ever reached the worker, it would take the same path, completed first and then refused a start.
• New worker integration test, using a new DuckShop edge case worker_freshness_skip.yml. One stale source has two dependent models; an independent model and a model on a fresh source come after them in execution order.

Verification

• On main the new test fails: the worker crashes on the rejected start, and the model on the fresh source ends skipped. With this change it ends success, and both stale-source models end skipped and never started.
• All 29 worker integration tests pass.
• make check-ci passes.

[kvlonge]

Folded into #1413 so the fixes share one CI run. The branch is kept.